Autonomous Car Following Models For Improving The Heterogeneous Traffic Stability

In the initial phase of CAVs implement in our country,according to the number which adding all levels of autonomous vehicles up,it’s obvious to say that the connected and autonomous vehicles is in a low market penetration rate,considering that it will take time to perfect the technology of connected and autonomous vehicles,it’s hard for connected and autonomous vehicle to achieve the explosive growth in a short time.On the other word,the connected and autonomous vehicle will share the public road resources with human driven vehicles(HDVs)for a period of time and the traffic will be in a mixed state.Adaptive cruising control(ACC)car following and cooperative adaptive cruising control(CACC)car following play important roles in connected and autonomous vehicles(CAVs),improving the traffic efficiency and traffic flow stability.Decision making processes are different between HDVs and AVs in car following events,and the difference will have an influence on the traffic stability.The human drivers’ car following character and autonomous vehicles car following character are considered respectively in this article,car following model for human drivers,autonomous car following model for adaptive cruising control and autonomous car following model for cooperative adaptive cruising control are proposed respectively considering the low market penetration of CAVs.The innovation and content of this study are included as following:(1)By combining the follower information of the objective HDV,proposing the extend car following model considering the follower chasing situation.By combining the LSTM with physical models based on the Physic-informed deep learning framework.The construction of the car following model mimicking the human driving is finished.According to the result of training the NGSIM dataset,the new model shows a better performance on the reflecting the decision made by human drivers than the model only considering the front vehicle information.The MSE descend by 9.38%.(2)Based on the analysis to car following data extracted from NGSIM dataset,the reward functions of deep reinforcement learning to improve the traffic stability are proposed.Constructing the virtual environment based on the Open AI gym to let autonomous driving Agent interact with it,DDPG algorithm is used to let Agent learn to follow the track in NGSIM,and autonomous driving model is obtained eventually.Constructing the reward function to improving the traffic stability and using it in virtual environment to encourage the Agent.The results show that the Agent can minimize 8%-16% of the acceleration fluctuation generated by the different classical physic car following model,thus the heterogeneous traffic stability is also improved.(3)For alleviating the degeneration of the CACC during the AVs under the low market penetration rate,avoiding its bad influence on the traffic stability.To simplify the previous constructed models of this study,a more convenient autonomous car following model is obtained for numerical simulation.Taking the classical car following model as the human car following model,a cooperative adaptive cruising control car following model is proposed considering the communication delay time,the human drivers’ reaction time under the specific information flow topologies.This model is combined with the previous models to form a mixed platoon,the stability of the platoon is proved with theoretical method and numerical simulation with python.The results show that the proposed model can relief from the deterioration of traffic stability brought by the degeneration of CACC,and improve the stability of the mixed platoon,so the heterogeneous traffic is more stable.The CAV car following model studied in this article,which aims to improve the stability of heterogeneous traffic flow,and provides the scientific basis for designing the single autonomous car following and cooperative autonomous car following respectively.Designed models are meaningful to improve the mixed traffic stability with considering the difference between HDVs and AVs.